Back in 2014, AMD announced that it would improve the power efficiency of its laptops by 25x by 2020. At the time, this seemed like an incredibly aggressive target, especially given the state of AMD’s Kaveri APU in 2014. Today, AMD announced that it broke its own target goal for energy efficiency. Compared with Kaveri, modern 7nm Ryzen Mobile CPUs are 31.7x more energy efficient by AMD’s estimation.
How’d the company do it? Let’s take a look. AMD’s methodology for calculating power efficiency starts with a 50/50 workload split between Cinebench R15 for CPU cores and 3DMark 11.
First, raw performance. In 2014, a 35W Kaveri FX-7600P scored 232 in Cinebench R15 (single-thread, multi-thread). Today, a Ryzen 4800H running at 35W TDP (aka, a 4800HS) turns in scores of 1727. In 3DMark 11, the 7600P scored 2142, while the 4800HS scores 5546. That’s an increase of 7.44x in Cinebench and 2.59x in 3DMark 11 while remaining within the same TDP factor. AMD claims a 5x performance improvement overall, which is what you get if you average the two individual improvements together.
According to AMD, it stuck with a 35W comparison target because that’s what it had used in 2014 when it had no 15W laptops.
The performance side is straightforward. The power efficiency gains have come from a wider range of places.
First of all, let’s acknowledge the elephant in the room. The largest share of AMD’s improvement comes from shifting to Carrizo from Kaveri. AMD did not disclose all the details of how it measured power efficiency, though this is fairly typical in the industry and the 25×20 initiative as a whole has been more transparent than we typically get.
AMD has said it is factoring in improvements like time-to-idle and that its overall efficiency metrics are weighted towards idle, which makes some sense, but that appears to be where the enormous 1.0 to 0.35 shift is coming from. From 2015 to 2019, AMD improved power efficiency in a modest stepwise fashion until 7nm allowed them to unleash much larger gains. Overall, the company is claiming to have improved its CPU power efficiency by 31.7x. That gain comes from multiplying the total gain in performance (measured in CB15 and 3DMark 11) against the improvement to power efficiency, which has improved by roughly 6x.
Energy Efficiency Doesn’t Necessarily Equate to Battery Life
One question this claim raises is why we haven’t seen a huge improvement in AMD’s battery life. If you don’t read laptop reviews on a regular basis, let me assure you, AMD laptops do not currently offer 12.5 days of battery life.
The Asus ROG Zephyrus has shown that AMD is capable of building a high-powered CPU with a Max-Q Nvidia RTX 2060, while still delivering 10-12 hours of battery life. That’s far better than any result AMD has ever turned in before, particularly when paired with high-end components. It wouldn’t surprise me in the slightest if median AMD battery life has at least doubled from Carrizo to Renoir.
The fact that AMD’s metrics are weighted towards idle and start with Kaveri is part of why the total efficiency gain is so high. If AMD had started counting with Carrizo, the gain would be a little more than 2x. Carrizo introduced some significant power efficiency improvements over Kaveri. As these Anandtech results show, in some cases, idle Carrizo power consumption is literally half of Kaveri’s idle power consumption. The gap isn’t always that large, but AMD’s claim of a threefold improvement from Kaveri to Carrizo doesn’t seem crazy. Neither do the further estimated gains from Carrizo through Renoir.
Another factor to keep in mind is that the CPU isn’t the only device drawing power in the system. The Wi-Fi card, display, RAM, and integrated storage all draw power, as does the onboard cooling solution. According to AMD, power consumption on Ryzen Mobile CPUs is now low enough that the CPU is not a majority of the total system power consumption. Which components a manufacturer chooses, however, still has a major impact on battery life. So does battery size — and manufacturers have been known to reduce battery capacity when more efficient CPUs become available, in order to use the space for other things.
AMD’s improvements to performance and power efficiency actually work against it in a specific way: Building faster APUs encourages OEMs to pair them with high-end components, including high-DPI displays, NVMe-attached storage, and larger RAM loadouts. All of these components have the capacity to draw far more power than the low-end components AMD systems shipped with a few years back.
In 2015, an AMD laptop might have 4-8GB of RAM in a single-channel DDR3L-1600 configuration, dual-channel if you were lucky. DDR4 was designed to use less power than DDR3L, but it also runs at higher RAM clocks. Between capacity improvements and clock increases, RAM might account for a larger amount of absolute power draw.
Making these kinds of improvements is what allows AMD to budget more power for these component upgrades in the first place, but it also means some of the additional battery life we’d get is effectively absorbed. Despite this, AMD battery life and its entire mobile position have improved dramatically in a short period of time.
Overall, well done AMD. The company’s 7nm processors have given it a commanding performance lead across a number of markets and price points, and the huge energy efficiency gains delivered over the past five years helped make that possible.
Now Read:
- What Kind of Performance Should We Expect From ARM-Based Macs?
- AMD Denies Rumors of Zen 3 Delay, Confirms Architecture On-Track for 2020
- AMD Announces Ryzen 3900XT, 3800XT, and 3600XT Performance Desktop CPUs
from ExtremeTechExtremeTech https://www.extremetech.com/computing/312117-amd-surpasses-its-own-25x20-goal-boosts-ryzen-power-efficiency-by-31x
No comments:
Post a Comment